The establishment of a monthly crop-mapping algorithm using high-resolution satellite images; for use in regions with highly variable agriculture
Abstract
Crop maps made by methods based on classifying satellite images provide useful information for farmers and policymakers. Historically, regions like Arequipa, Peru have challenged such methods due to small fields (often less than 0.1 ha), a lack of growing season patterns, and variable land management practices. Advances in remote sensing and computational technology have expanded classification possibilities to include regions like Arequipa. However, there is still a lack of case studies and general methods for the classification of such regions. This study aimed to establish a crop-mapping algorithm (CMA) for the city of Arequipa, Peru. Monthly ground reference data were collected near the city of Arequipa from September 2019 to February 2020 using the app Epicollect5. The data input to the CMA are filtered to ensure that the crops are of sufficient age to be classified. The data are then used to test and validate classification methods: for each month, the RandomForest, k-nearest-neighbor, and maximum likelihood methods are tested and the one producing the best validation accuracy is chosen. The CMA uses the chosen methods to classify monthly high-resolution PlanetScope satellite images, then uses unique spatial- and temporal-correction algorithms to make the maps consistent in space and time. The CMA randomly samples training and validation data from the filtered data to iteratively perform the classification and correction process, then averages the iteration results. The CMA outputs: confidence intervals for mean validation accuracies and mean cover statistics; and maps that best fit the mean results. The CMA had generally high accuracies for six months and ten cover types (mean monthly accuracies usually greater than 70%). The correction algorithms significantly improved the accuracies and spatial and temporal consistencies of the maps. The maps have high spatial (3.0m) and temporal (monthly) resolution, allowing for a detailed look at the agriculture of the region.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2020
- Bibcode:
- 2020AGUFMH122...06D
- Keywords:
-
- 1655 Water cycles;
- GLOBAL CHANGE;
- 1807 Climate impacts;
- HYDROLOGY;
- 1813 Eco-hydrology;
- HYDROLOGY;
- 1880 Water management;
- HYDROLOGY